How to Ask for Permission

Adrienne Porter Felt, Serge Egelman, Matthew Finifter, Devdatta Akhawe, David A. WagnerHotSec 2012 · OpenReview

TL;DR

Modern applications require access to sensitive user data and device functionalities, leading to complex permission systems. This paper proposes a novel approach where platforms select the most appropriate permission-granting mechanism for each individual permission, rather than using a one-size-fits-all approach. This aims to reduce user habituation to warnings and improve the overall user experience with security and privacy controls.

What Problem Does This Paper Address?

As applications become more integrated with our devices and personal lives, they increasingly need access to sensitive resources like location, contacts, and camera. Permission systems are designed to mediate this access, informing users and allowing them to make decisions. However, current systems, particularly on popular smartphone platforms like Android and iOS, present a wide disparity in user experience. Some bombard users with install-time warnings, leading to habituation and a lack of attention, while others offer very limited runtime controls. This creates a challenge for platform designers: how to effectively balance security and privacy with a user experience that doesn't overwhelm or frustrate users. The paper addresses the lack of a principled, granular approach to selecting permission-granting mechanisms.

Methods

The authors propose a framework for platform designers to choose the most suitable permission-granting mechanism for each permission. This involves a decision procedure guided by two core principles: conserving user attention (only asking for consent when consequences are severe) and avoiding interruptions (integrating security decisions into the user's natural workflow where possible).

They identify and analyze four primary permission-granting mechanisms:

  • Automatic Grant: Permissions are granted without explicit user action, with mechanisms for auditing and undoing. Suitable for low-severity or easily reversible actions.
  • Trusted UI: Permission requests are embedded within the application's workflow (e.g., photo choosers, review screens, embedded buttons). This is non-interruptive and leverages user motivation.
  • Runtime Consent Dialogs: Traditional pop-up dialogs that interrupt the user's task to ask for permission. Used when other mechanisms are not feasible.
  • Install-Time Warnings: Permissions are listed during application installation. This is the least desirable due to its interruptive and repetitive nature.

A decision tree (Figure 1) guides the selection process, prioritizing automatic grants and trusted UI over runtime dialogs and install-time warnings.

Datasets

The paper mentions a survey of 3,115 smartphone users to identify low-severity permissions. They also conducted a preliminary evaluation by manually testing 20 popular iPhone applications to assess how their proposed system would impact user experience. No new datasets were introduced for training or benchmarking.

Key Insights and Ideas

  • Permission Granularity is Key: The core idea is that not all permissions are created equal. Applying a single mechanism to all permissions is inefficient and leads to poor user experience.
  • Habituation is a Real Threat: Repetitive warnings, even for important permissions, lead to users clicking through without reading, rendering the warnings ineffective. This is a major driver for prioritizing non-interruptive mechanisms.
  • User Attention is Finite: Security mechanisms must be designed with the understanding that users have limited cognitive resources. Interruptions and excessive prompts deplete this resource, leading to satisficing behavior.
  • Trusted UI Offers a Sweet Spot: Integrating permission requests into the natural flow of an application (Trusted UI) is highly effective because it leverages user motivation and expectations.
  • Auditing is Crucial for Automatic Grants: For permissions granted automatically, robust auditing and undo mechanisms are essential to prevent abuse and empower users.

Key Results

  • The proposed model suggests that a significant majority of permissions can be handled with non-interruptive mechanisms: 55% can be automatically granted, and 23% can be represented with trusted UI.
  • Only 16% would require runtime consent dialogs, and a mere 6% would necessitate install-time warnings.
  • A preliminary evaluation on iPhone applications indicated that the proposed system would decrease the average number of runtime consent dialogs per application (from iOS's current average to 0.25).
  • The average application would contain 0.60 trusted UI elements, with only a few requiring minor changes to adopt trusted UI elements.
  • These results suggest the approach could plausibly provide a workable permission-asking system without triggering habituation.

Limitations and Future Work

  • Platform Neutrality vs. Implementation: While the paper proposes platform-neutral permissions, actual implementation may require API changes, which can be complex.
  • Developer Incentives: The paper acknowledges that developers might choose suboptimal mechanisms if not properly incentivized, potentially negating the benefits of a well-designed system.
  • Comprehensive Usability Evaluation: The current evaluation is preliminary. The authors suggest further research is needed to perform a comprehensive usability evaluation of a platform that fully adopts their model.
  • Application Review Process: The work is primarily targeted at designers of new platforms. Platforms with existing application review processes might benefit differently.

Summary updated

BibTeX
@inproceedings{felt2012how,
  author = {Adrienne Porter Felt and Serge Egelman and Matthew Finifter and Devdatta Akhawe and David A. Wagner},
  title = {{How to Ask for Permission}},
  booktitle = {HotSec},
  year = {2012},
  url = {https://openreview.net/forum?id=thQiQklaex},
}
Figure